Extract Text (Tesseract OCR)

@tesseract1 VersionExtract

Runs the classic CPU Tesseract engine over each Image, emitting either a joined String transcript or per-word regions with Rectangle<Double> bounds and optional confidence normalised to 0..1.

How it fits

Typical backends

OCR scanned pages from files and deliver the transcript over HTTP.

Crop a region of interest, OCR it, and summarise the text with a text model.

Caveats

I/O contractThe output arm is fixed when the backend deploys: a text consumer pins the String transcript, a region consumer pins per-word records with Rectangle<Double> bounds, optionally carrying a confidence score. Tesseract's native word box is an axis-aligned rectangle.
Hard constraintlanguage must name a Tesseract language data pack installed in the image; a missing pack aborts the call.
Parameter interactiontesseract_psm selects the page-segmentation mode, which strongly affects results on sparse or single-line images, and tesseract_oem selects the engine mode.
LimitationThis is a CPU-only engine with no GPU path; throughput is bounded by CPU. Words with negative confidence are dropped before emission.
AccuracyTesseract is sensitive to skew and low resolution; deskew and upscale faint scans upstream for reliable reading.
State lifetimeLanguage and mode settings are read once at startup; changing them requires a redeploy.

Versions

df477d1clatestdefault
Automated release
6/29/2026